JP2009251779A - Sales estimation system, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically estimate demands changing according to the lapse of time, that is, the number of sales. <P>SOLUTION: The sales estimation system includes: a sales database; a short term data group generation part for generating the set of short term data by using attributes as sales conditions, and using a label as a first label when one merchandise or service is sold and the label as a second label otherwise; a long term data generation part for generating long term data; a long term classifier generation part for generating a classifier to output scores as a real number value showing first label likeness or second label likeness as a rule for determining a label based on the attributes of long term data; a short term threshold group estimation part for calculating the scores of short term data by using the classifier, and for calculating the threshold of scores; a time series analyzing part for analyzing the time series change in a threshold; a virtual short term data generation part for generating virtual short term data; a score calculation part for calculating scores; a threshold estimation part for estimating a threshold on an estimation object time; and a sales estimation part for estimating the number of sales on the estimation object time. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a sales number prediction system, a sales number prediction method, and a sales number prediction program for predicting demand for goods or services sold by retailers, and in particular, a sales number prediction system for predicting demand using customer information, and the number of sales. The present invention relates to a prediction method and a sales volume prediction program.

  In a retail store, a POS (Point of Sales) system is used, and the sales data of the collected products is used in various ways. Among them, there is a system that predicts demand for goods based on sales data and uses it for ordering.

  For example, Patent Literature 1 describes a system that predicts demand for a product based on sales results of the product, taking into account fluctuations in sales due to various fluctuation factors such as selling price and weather.

  Patent Document 2 describes a method for forecasting sales for each unit specification constituting a new product. The sales composition ratio for each unit specification is calculated by substituting the product characteristics of the product into the sales composition ratio model function expressed by a neural network formula.

  Patent Document 3 describes an apparatus for predicting a sales amount from a past sales performance quantity. This is a method of systemizing rules of thumb possessed by experts and forecasting sales using the rules. In order to verify the adjusted results, simulation is performed with virtual data manually input and alarms are displayed. Means are also described.

  Patent Document 4 describes a system for grasping the tendency of simultaneous sales of consumers from analysis of POS data.

JP 2004-334326 A JP 2003-122895 A Japanese Patent Laid-Open No. 07-175787 JP 2007-094592 A

  The value of goods and services changes with the passage of time, and the demand changes accordingly. However, the demand prediction described in Patent Document 1 cannot predict the demand that changes with the passage of time. The reason is that, on the assumption that the demand does not change with the passage of time, the demand is predicted by adjusting the long-term sales performance by the influence of various fluctuation factors.

  The sales number prediction system described in Patent Documents 2 and 3 cannot automatically predict the sales number without knowledge from an expert. The reason is that Patent Document 2 does not disclose a method for automatically estimating product characteristics, and in the embodiment, both the current car and the new car are configured so that developers specify the characteristic values in percentage. Because. Therefore, it is necessary to quantify product characteristics by an expert called a developer, and the number of sales cannot be automatically predicted. Patent Document 3 also does not disclose a method for automatically predicting a sales amount without an expert rule of thumb. Production rules and membership functions are registered in advance in the fuzzy neuro data table, and a method for automatically generating the production rules and membership functions is not disclosed. Therefore, a production rule for predicting a demand that changes with the passage of time needs to be given by an expert, and can only be analyzed within a range of rules known by the expert.

  The marketing data collection and analysis system described in Patent Document 4 can analyze the number of sales, but cannot predict it.

  Moreover, although the value of goods and services changes according to the characteristic of an individual customer, patent documents 1, 2, and 3 do not consider such a difference. The reason is that processing is performed based on the number of sales, and differences among individual customers are summed and offset. In the system described in Patent Document 4, customer characteristics are taken into account, but it is the user of the system that designates the customer characteristics, and demand according to the automatically classified customer characteristics cannot be analyzed.

  Therefore, an object of the present invention is to provide a sales number prediction system, a prediction method, and a prediction program for automatically predicting the demand for goods or services that change over time, that is, the number of sales.

  The sales number prediction system according to the present invention has a sales database that stores information indicating one or more sales conditions and past sales of a plurality of products or services, and attributes based on the information stored in the sales database. And a label is a first label if one product or service is sold, and a second label if the product or service is not sold, and a set of short-term data that is a set of attributes and labels in a short time series is generated. Based on the short-term data group generation unit, the short-term data group generated by the short-term data group generation unit, the long-term data generation unit that generates long-term data including all the short-term data, and the label is determined based on the attributes of the long-term data A long-term classifier generator for generating a classifier that outputs a score that is a rule and is a real value indicating the likelihood of the first label or the second label; A short-term threshold group estimation that calculates a score of each short-term data using a container, and calculates a threshold value of a score corresponding to the boundary between the first label and the second label based on the label of each short-term data and the calculated score Virtual short-term data including virtual sales conditions as attributes based on attributes of long-term data and sales conditions at the prediction target time Based on the virtual short-term data generation unit, the virtual short-term data generated by the virtual short-term data generation unit and the classifier generated by the long-term classifier generation unit, the score calculation unit that calculates the score, and the time series analysis unit analyzed Based on the result, a threshold estimation unit that estimates a threshold at the prediction target time, a score calculated by the score calculation unit, and a threshold estimated by the threshold estimation unit are used to predict the number of sales at the prediction target time. Characterized by comprising a sales prediction unit.

  The method for predicting the number of sales according to the present invention is based on information stored in a sales database storing information indicating one or more sales conditions and past sales of a plurality of products or services. If a single product or service is sold, the first label is generated, and if it is not sold, a short-term data set that is a set of attributes and labels in a short time series is generated and generated. A rule that generates long-term data including all short-term data based on a set of data and determines a label based on attributes of the long-term data, and a score that is a real value indicating the likelihood of the first label or the second label A classifier to be output is generated, a score of each short-term data is calculated using the classifier, and the first label and the second label are calculated based on the label of each short-term data and the calculated score. Calculate the threshold of the score corresponding to the boundary with the bell, analyze the time-series change of the threshold, and include virtual sales conditions as attributes based on long-term data attributes and sales conditions at the forecast target time Generate virtual short-term data, calculate the score based on the generated virtual short-term data and the classifier, estimate the threshold at the prediction target time based on the analysis result, and calculate the calculated score and the estimated threshold Based on this, the number of sales at the prediction target time is predicted.

  The sales number prediction program according to the present invention uses an attribute as a sales condition based on information stored in a sales database that stores information indicating one or more sales conditions and past sales of a plurality of products or services in a computer. , A short-term data set that is a set of attributes and labels in a short time series in a short time series, with the label as the first label if one product or service is sold and the second label otherwise. Rules for determining labels based on long-term data attributes and long-term data generation processing that generates long-term data that includes all short-term data based on a set of short-term data generated by data group generation processing and short-term data group generation processing A long-term classifier generation process for generating a classifier that outputs a score that is a real value indicating the likelihood of the first label or the second label Calculating a score of each short-term data using a classifier, and calculating a threshold of a score corresponding to the boundary between the first label and the second label based on the label of each short-term data and the calculated score Based on group estimation processing, time-series analysis processing that analyzes time-series changes in thresholds, and long-term data attributes and sales conditions at the forecast target time, virtual short-term data that includes virtual sales conditions as attributes Based on the virtual short-term data generation process to be generated, the virtual short-term data generated by the virtual short-term data generation process, and the classifier generated by the long-term classifier generation process, a score calculation process that calculates a score, and a time series analysis process Based on the analysis result, the threshold estimation process for estimating the threshold at the prediction target time, the score calculated by the score calculation process, and the threshold estimated by the threshold estimation process, Characterized in that to execute a sales prediction process of predicting the number of sales target time.

  According to the present invention, there is an effect that it is possible to predict the demand for goods or services that change with the passage of time, that is, the number of sales, in consideration of the fluctuation factors.

  First, an outline of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a minimum configuration example of a sales number prediction system according to the present invention. The sales number prediction system illustrated in FIG. 1 includes a sales database 1, a short-term data group generation unit 31, a long-term data generation unit 32, a long-term classifier generation unit 33, a short-term threshold group estimation unit 34, and a time series analysis. Unit 35, virtual short-term data generation unit 41, score calculation unit 42, threshold estimation unit 43, and sales number prediction unit 44.

  The sales database 1 stores information indicating one or more sales conditions and past sales of a plurality of products or services.

  Based on the information stored in the sales database 1, the short-term data group generation unit 31 sets the attribute as the sales condition, sets the label as the first label if one product or service is sold, and sets the label as the second label otherwise. A short-term data set that is a set of attributes and labels in a short time series in a time series is generated.

  The long-term data generation unit 32 generates long-term data including all the short-term data based on the set of short-term data generated by the short-term data group generation unit 31.

  The long-term classifier generation unit 33 is a rule for determining a label based on attributes of long-term data, and generates a classifier that outputs a score that is a real value indicating the likelihood of the first label or the second label.

  The short-term threshold value group estimation unit 34 calculates a score of each short-term data using a classifier, and based on the label of each short-term data and the calculated score, a score corresponding to the boundary between the first label and the second label The threshold value is calculated.

  The time series analysis unit 35 analyzes a time series change of the threshold value.

  The virtual short-term data generation unit 41 generates virtual short-term data including virtual sales conditions as attributes based on long-term data attributes and sales conditions at the prediction target time.

  The score calculation unit 42 calculates a score based on the virtual short-term data generated by the virtual short-term data generation unit 41 and the classifier generated by the long-term classifier generation unit 33.

  The threshold estimation unit 43 estimates the threshold at the prediction target time based on the result analyzed by the time series analysis unit 35.

  The sales number prediction unit 44 predicts the number of sales at the prediction target time based on the score calculated by the score calculation unit 42 and the threshold value estimated by the threshold value estimation unit 43.

  With the configuration shown in FIG. 1, it is possible to predict the demand for goods or services that change with the passage of time, that is, the number of sales, while taking into account the fluctuation factors.

  A preferable aspect of the sales number prediction system according to the present invention is based on, for example, a sales database storing information indicating one or more sales conditions and past sales of a plurality of products or services, and information on the sales database. The attribute is the sales condition, the label is the first label if one product or service is sold, and the second label if it is not sold. A short-term data group generation unit for generating a set, a long-term data generation unit for generating long-term data based on a set of short-term data, and a rule for determining a label from attributes of long-term data, which is the first label-likeness ( Or a long-term classifier generator that generates a classifier that outputs a score that is a real value of the second label likelihood), and a classifier calculates a score of each short-term data, Based on the label and score of the period data, the short-term threshold group estimator that calculates the threshold of the score, the time series analyzer that analyzes the time-series change of the threshold, and the attribute generated by the long-term data Score is calculated based on the virtual short-term data generation unit that generates virtual short-term data, the virtual short-term data and the classifier, based on the virtual data attributes that are virtual sales conditions and the sales conditions of the forecast time. Based on the analysis by the score calculation unit, the time series analysis unit, the threshold estimation unit for estimating the threshold of the prediction target time, and the sales of the prediction target time based on the score and the threshold calculated by the score calculation unit A sales number prediction unit for predicting the number.

  Another preferable aspect of the sales number prediction system according to the present invention includes, for example, a sales database storing information indicating one or more sales conditions and past sales of a plurality of products or services, and information of the sales database. A short term that is a pair of attributes and labels in a short time series, with the attribute as the sales condition, the label as the first label if one product or service is sold, and the second label if it is not sold A short-term data group generation unit that generates a data set, a long-term data generation unit that generates long-term data based on the short-term data set, and a rule for determining a label from attributes of long-term data, the first label A long-term classifier generator that generates a classifier that outputs a score that is a real value of the likelihood (or second label likelihood), and the classifier calculates the score of each short-term data Based on each short-term data label and score, a short-term threshold group estimator that calculates a threshold for the score, and a virtual data score based on the virtual data attribute and classifier generated based on the long-term data attribute Calculating a virtual data label based on the score and the threshold estimated by the short-term threshold value group estimation unit, and calculating a sales rate based on the determined label; a sales rate and a threshold; Based on a threshold function estimation unit that estimates a function of the virtual data, a virtual short-term data generation unit that generates virtual short-term data based on the attribute of the virtual data and the sales condition of the prediction target time, a virtual short-term data and a classifier Based on the score calculation unit for calculating the score, the threshold value estimation unit for estimating the threshold value of the prediction target time based on the analysis in the time series analysis unit, and the score and threshold value calculated by the score calculation unit Characterized in that it comprises a sales prediction unit that predicts the number of sales prediction target time.

  Further, another preferable aspect of the sales number prediction system according to the present invention is, for example, in addition to one of the two configurations described above, a customer database that stores information on one or more customers, and information on sales database information. Some include a sales database that stores correspondence with customers, and the short-term data group generation unit is characterized in that the attributes are sales conditions and customer information.

Embodiment 1. FIG.
A first embodiment of the present invention will be described below with reference to the drawings. Not only those who use products and services for a fee, but also those who use products and services for free, and potential customers who are likely to use products and services in the future are called customers. If the product or service is free, the selling price is 0. In addition, the concept of goods or services includes not only finished products such as products but also minimum units for maintaining the functions of the products. Furthermore, a certain product or service category can be considered as one product and applied.

  The sales number prediction system according to the present invention predicts time-series changes in sales numbers. FIG. 2 is a block diagram showing a configuration example of the first embodiment of the sales number prediction system according to the present invention. The sales number prediction system according to the first embodiment shown in FIG. 2 includes a sales database (hereinafter referred to as a sales DB) 1 that stores data of sales results, a storage unit 2, and sales results while taking into account various fluctuation factors. Are provided with an analysis unit 3 for analyzing the above and a prediction unit 4 for predicting the number of sales.

  The sales DB 1 stores information (sales result data) indicating one or more sales conditions and past sales of a plurality of products or services. The sales DB 1 stores data obtained by collecting information on products or services sold at one cash register as sales performance data. FIG. 3 is an explanatory diagram illustrating an example of data stored in the sales DB 1. The sales record data stored in the sales DB 1 includes any one or more of various information such as the selling price of the product or service, the number of sales, the weather, the day of the week, and the time. FIG. 3 illustrates a case where the sales DB 1 stores data including the selling price of the products A and B, the presence / absence of supply, and the number of sales in association with the sales conditions. The sales condition is information including, for example, the date, day of the week, time of day, weather, and the like when the product is sold, and is information indicating a condition for selling the product. Further, the data stored in the sales DB 1 can include information on other products or services as well as products or services for which the number of sales is to be predicted.

  The storage unit 2 includes a long-term database (hereinafter referred to as long-term DB) 21, a condition database (hereinafter referred to as condition DB) 22, a classifier database (hereinafter referred to as classifier DB) 23, and a threshold database (hereinafter referred to as “classifier DB”). , Referred to as threshold DB).

  The long-term DB 21 stores long-term data generated by a long-term data generation unit 32 described later.

  The condition DB 22 stores a set of sales conditions in a certain period and the number of times of sale that is the number of times of sale (number of cashiers) for a customer who has visited the store. Here, when a plurality of products and services are sold simultaneously, the number of sales is one.

  The classifier DB 23 stores a rule (classifier) for outputting a real value indicating the ease of purchasing a certain product or service.

  The threshold value DB 24 stores an expression for obtaining a threshold value for determining whether or not to purchase for each product or service in association with time.

  The analysis unit 3 includes a short-term data group generation unit 31, a long-term data generation unit 32, a long-term classifier generation unit 33, a short-term threshold group estimation unit 34, and a time series analysis unit 35.

  For example, when a plurality of periods (a set of periods) is input by the user, the short-term data group generation unit 31 generates a plurality of short-term data corresponding to the plurality of periods based on the sales DB 1. Here, the term set is a set of periods to be analyzed in the period in which the sales record is input to the sales DB 1. For example, if there are sales results for 90 days and the period is one day at a time, the set of periods is input as the first day, the second day,. The input period may be 3 hours or 1 week.

  The short-term data is information including an attribute and a class (label). The attribute is, for example, information including the sales conditions included in the sales performance data stored in the sales DB 1, the selling price of each product, and the presence / absence of supply. The class is information indicating whether the product has been sold. Hereinafter, short-term data in a predetermined period (time) t may be expressed as short-term data t.

  FIG. 4 is an explanatory diagram showing an example of short-term data when the sales number is predicted for the product A. FIG. 4 illustrates a case where the period is March 10, 2008 and the attributes (items) of the short-term data are used except for the number of sales of the target product A and the number of sales of other products. The short-term data group generation unit 31 generates short-term data by setting the class as positive (+) if the number of sales of the target product A is 1 or more, and setting the class as negative (-) if the number is 0 or less.

  In addition, the short-term data group generation unit 31 stores a set of sales conditions and the number of sales in a predetermined period in the condition DB 22. The short-term data group generation unit 31 extracts, for example, sales performance data including sales dates that match the input period, calculates the number of sales, and stores it in the condition DB 22.

  The long-term data generation unit 32 generates long-term data by combining all the short-term data generated by the short-term data group generation unit 31 and stores the long-term data in the long-term DB 21.

  The long-term classifier generator 33 generates a classifier based on the long-term data generated by the long-term data generator 32 and stores the classifier in the classifier DB 23.

  The classifier according to the present invention is a rule that outputs a score which is a numerical value indicating the likelihood of a first label (or the likelihood of a second label) when data that is a set of an attribute and a class (label) is input. For example, the first label is information indicating that a product or service has been sold, and the second label is information indicating that the product or service has not been sold. That is, the classifier is a rule in which a real value (score) indicating the ease of purchasing a product is associated with a set of an attribute and a class, for example. A method by which the long-term classifier generator 33 generates a classifier will be described later.

  The short-term threshold value group estimation unit 34 estimates (calculates) a plurality of threshold values based on the classifier generated by the long-term classifier generation unit 33 and a plurality of short-term data. The threshold value calculated by the short-term threshold value group estimation unit 34 is a score threshold value, for example, a score value corresponding to a boundary between a positive (+) class and a negative (−) class.

  The time series analysis unit 35 performs time series analysis on a plurality of threshold values estimated by the short-term threshold value group estimation unit 34, and stores a formula for obtaining the threshold values in the threshold value DB 24. That is, the time series analysis unit 35 analyzes the time series change of the threshold value.

  The prediction unit 4 includes a virtual short-term data generation unit 41, a score calculation unit 42, a threshold value estimation unit 43, and a sales number prediction unit 44.

  For example, when a prediction target period (time) and a sales condition in the prediction target period are input by the user, the virtual short-term data generation unit 41, based on the long-term DB 21 and the condition DB 22, for example, An attribute of virtual short-term data indicating an attribute that is a sales condition is generated. The virtual short-term data is, for example, virtual sales data for predicting the number of sales.

  The prediction target period is information indicating a target period for which the user wants to predict the sales volume, for example. The sales condition in the prediction target period is information indicating a condition for selling a product in the prediction target period, for example.

  The score calculation unit 42 calculates a score for the individual data (each record) of the virtual short-term data based on the classifier stored in the classifier DB 23 and the attributes of the virtual short-term data.

  The threshold value estimation unit 43 estimates a threshold value in the prediction target period based on an expression for obtaining a threshold value stored in the threshold value DB 23.

  The sales number prediction unit 44 predicts the number of sales in the prediction target period based on the score of the individual data of the virtual short-term data and the threshold value of the prediction target period.

  It should be noted that a set of analysis target periods input to the short-term data group generation unit 31 of the analysis unit 3, a sales period for the prediction target period and the prediction target period input to the virtual short-term data generation unit 41 of the prediction unit 4, etc. May be input via an input device (not shown) such as a keyboard. Further, an output device (not shown) such as a display for outputting the predicted number of sales may be provided.

  The sales number prediction system can be realized by a computer, and the analysis unit 3 and the prediction unit 4 constituting the sales number prediction system are realized as a program for causing a computer processing device (CPU) to realize the above-described functions. Is possible. The computer may incorporate the sales DB 1 and the storage unit 2.

  Next, the operation of the first embodiment will be described with reference to the drawings.

  FIG. 5 is a flowchart showing an example of processing progress of the analysis unit 3 in the sales number prediction system of the present invention. For example, when T sets of periods are input via an input device or the like, the sales number prediction system operates as follows.

  The short-term data group generation unit 31 creates short-term data 1,..., Short-term data T, which are short-term sales data, based on the sales DB1 and the set period input (step S101). Further, the short-term data group generation unit 31 stores a set of the sales condition and the number of sales in the period in the condition DB 22 (step S102). Here, from the sales conditions, conditions such as the selling price of a product, which are not known before the store visit and are considered to be affected after the store visit, are excluded. The number of sales is the number of individual data (each record) of short-term data.

  The long-term data generation unit 32 creates the long-term data by combining the short-term data 1,..., The short-term data T, and stores it in the long-term DB 21 (step S103).

  The long-term classifier generation unit 33 generates a classifier using long-term data and stores it in the classifier DB 23 (step S104).

  The classifier according to the present invention is a rule that outputs a score that is a numerical value of first label likelihood (or second label likelihood) when data that is a combination of an attribute and a class is input. The long-term classifier generation unit 33 generates a classifier using various algorithms for creating a classifier when data that is a set of attributes and classes is input.

  For example, the classifier may be any classifier whose output is a continuous value, such as a neural network, a support vector machine, or ensemble learning, or a combination thereof. Here, for convenience of explanation, it is assumed that the long-term classifier generation unit 33 generates a classifier whose output value (score) is large when the class is positive.

  The short-term threshold value group estimation unit 34 first initializes the time t to 1 (step S105). Next, the short-term threshold value group estimation unit 34 calculates the score of the short-term data t using the classifier generated by the long-term classifier generation unit 33 (step S106). For example, the short-term threshold value group estimation unit 34 causes the classifier stored in the classifier DB 23 to input the attribute and class included in the short-term data t and output the score.

  Then, the short-term threshold value group estimation unit 34 calculates a score threshold value based on the class and score of the short-term data t. For example, the short-term threshold value group estimation unit 34 predicts a class based on a score (threshold value) while changing a threshold value that is a real number, calculates an error based on the predicted class and the actual class, and minimizes the error. Is obtained (hereinafter referred to as Thres [t]) (step S107). The error is, for example, a number that is different between the predicted class and the actual class. Alternatively, the absolute value of the difference between the number of individual data whose actual class is positive and the number of individual data whose predicted class is positive.

  FIG. 6 is an explanatory diagram illustrating an example of the score of the short-term data t. In step S106, the short-term threshold value group estimation unit 34 calculates a score indicating ease of purchase for individual data in the short-term data, as shown in FIG. In the example illustrated in FIG. 6, the short-term threshold value group estimation unit 34 predicts a class for the threshold value 0.6 and a class for the threshold value 0.8 in step S107, respectively. The short-term threshold value group estimation unit 34 calculates the error between the class for the threshold value 0.6 and the actual class as 0 for the individual data of ID1, and the class and actual value for the threshold value 0.8 for the individual data of ID2. The error from the class is calculated as 1. In the example illustrated in FIG. 6, for example, assuming that one short-term data is three individual data items ID1 to ID3, the error at the threshold 0.6 is 0 and the error at the threshold 0.8 is 1. The short-term threshold value group estimation unit 34 calculates Thres [t] = 0.6.

  The short-term threshold value group estimation unit 34 compares t with T (step S108). If t <T (Yes), the operation moves to step S109. If t ≧ T (No), the operation moves to step S110. .

  In step S109, the short-term threshold value group estimation unit 34 increments t and moves the operation to step S106.

  The time series analysis unit 36 estimates a formula for obtaining a threshold value Thres [t] (t = 1,..., T) with respect to the time t by time series analysis and stores the estimated value in the threshold value DB 24 (step S110). The time series analysis unit 36 estimates, for example, an autoregressive moving average model as represented by the following expression by time series analysis.

  Here, a [i] and b [j] are arbitrary constant parameters, p is the order of autoregression, q is the order of the moving average of past residuals, and e [t] is the residual. . In the estimation of the model, assuming that e [t] is a normal distribution with an average of 0, various parameters may be estimated so as to match the data Thres [t] (t = 1,..., T). Here, before the processing, an outlier of the threshold may be appropriately complemented and used.

  FIG. 7 is a flowchart showing an example of processing progress of the prediction unit 4 in the sales number prediction system of the present invention. The prediction unit 4 operates as follows when a prediction target period (prediction target time and its period), a selling price in the prediction target period, and sales conditions such as weather are input via a keyboard input device or the like. .

  The virtual short-term data generation unit 41 first extracts an attribute similar to the input sales condition from the long-term DB 21 and rewrites the sales condition so as to match the input sales condition in the prediction target period (time). Virtual short-term data is set (step S201). Here, the virtual short-term data generation unit 41 may extract an attribute after adding an attribute generated by randomly changing a part of the attribute of the long-term data.

  The virtual short-term data generation unit 41 sets the number of individual data in the virtual short-term data to NALL (step S202). That is, NALL is the number of sales in the virtual short-term data.

  The score calculation unit 42 calculates the score of the virtual short-term data from the attribute of the virtual short-term data using the classifier stored in the classifier DB 23 (step S203).

  The threshold value estimation unit 43 calculates a threshold value in the prediction target period based on an expression for obtaining a threshold value stored in the threshold value DB 24 (step S204).

  The sales number prediction unit 44 calculates the number of individual data NP predicted to be sold based on the predicted label and the threshold (step S205). For example, the sales number prediction unit 44 compares the score of the virtual short-term data calculated in step S203 with the threshold calculated in step S204, and the number of individual data of the virtual short-term data predicted to be positive (sales) in the class NP is calculated.

  Then, the sales number prediction unit 44 extracts the number N of sales that matches the input sales conditions from the condition DB 22 (step S206), and sets the number of individual data predicted to be sales as N × NP / NALL or the like. Normalization is performed to obtain the predicted sales number of the target product or service (step S207).

  Here, when there are a plurality of conditions (number of sales) that match the input sales conditions in the condition DB 22, for example, the average of the plurality of sales times is used. Or it may be the number of recent sales. If the condition DB 22 does not have a condition that matches the input sales condition, the number of times of sales under the most similar condition is set.

  Moreover, although the example which outputs only the estimated sales number was demonstrated, in order for the user of a system to know reliability, you may output information, such as the error which the analysis part 3 calculated. For example, a threshold graph estimated by the short-term threshold group estimation unit 34 or an error that is a difference between the threshold estimated by the short-term threshold group estimation unit 34 and the threshold estimated by the time series analysis unit 35 is output to the output device. May be.

  Moreover, you may produce | generate data, a classifier, etc. from sales DB1, as needed, without providing the memory | storage part 2. FIG.

  In addition, although the example in which the prediction unit 4 predicts the number of sales following the analysis by the analysis unit 3 has been described, the analysis can be performed by the analysis unit 3 alone. Moreover, you may operate the prediction part 4 based on the result which the analysis part 3 analyzed beforehand. Further, the prediction unit 4 may be operated after the analysis result by the analysis unit 3 is corrected or supplemented by the user of the system.

  As described above, according to the first embodiment, a classifier is generated using long-term data including various fluctuation factors, and the short-term data is predicted, so that a time-series change is represented by a change in threshold value. Estimate the number of sales. Accordingly, it is possible to predict a time-series change in the number of sales while taking into account various fluctuation factors. Also, a score of ease of purchase under individual sales conditions is calculated. Therefore, it is possible to analyze optimal sales conditions.

Embodiment 2. FIG.
The sales number prediction system of the second embodiment is a system including a customer database in addition to the configuration of the first embodiment, and can analyze and predict customer characteristics and sales conditions that are easy to purchase. That is, the sales number prediction system of the second embodiment aims to predict the demand of individual customers according to automatically classified customer characteristics.

  FIG. 8 is a block diagram showing a configuration example of the second embodiment of the sales number prediction system according to the present invention. The sales number prediction system according to the second embodiment includes a sales DB 1, a customer database (hereinafter referred to as a customer DB) 5, data including customer information, a storage unit 2, an analysis unit 6, and sales numbers. And a prediction unit 4 that predicts.

  The sales DB1, the storage unit 2, and the prediction unit 4 are the same as the sales DB1, the storage unit 2, and the prediction unit 4 in the first embodiment, respectively. In the analysis unit 6, the short-term data group generation unit 31 is changed to a short-term data group generation unit 61 with respect to the analysis unit 3 in the first embodiment. Hereinafter, only different parts from the first embodiment will be described.

  The customer DB 5 stores data obtained by collecting customer characteristics by operating a customer card or a salesclerk at the time of sale. FIG. 9 is an explanatory diagram illustrating an example of data stored in the customer DB 5. The data stored in the customer DB 5 includes the age, sex, occupation, etc. of the customer, the day / time and purchase frequency most frequently purchased in the entire product, and the day / time purchased most frequently in various product or service categories. It includes any one or more of various information such as purchase frequency. In FIG. 9, the customer DB 5 associates the customer ID with the customer ID that can identify the customer, and indicates the most frequent day, the most frequent time, and the frequency at which the customer's characteristics such as age, sex, occupation, and product are most proactively purchased. The case where the information containing is memorize | stored is illustrated.

  In addition, sales DB1 may memorize | store the information (for example, customer ID) which can be matched with a customer.

  For example, when a plurality of periods are input by the user, the short-term data group generation unit 61 generates a plurality of short-term data corresponding to the plurality of periods based on the sales DB 1 and the customer DB 5. For example, the short-term data group generation unit 61 uses the customer's age, gender, occupation, day of the week / time of purchase most frequently in the entire product, purchase frequency, and most of the various product or service categories as the customer information as attributes. Short-term data including any information indicating the day and time of frequent purchase, purchase frequency, initial purchase time, and final purchase time is generated.

  Next, the operation of the second embodiment will be described with reference to the drawings.

  FIG. 10 is a flowchart showing an example of processing progress of the analysis unit 6 in the sales number prediction system of the present invention. For example, when T sets of periods are input via an input device or the like, the sales number prediction system operates as follows.

  The short-term data group generation unit 61 generates a plurality of short-term data corresponding to a plurality of periods based on the sales DB 1 and the customer DB 5 (step S301).

  11 is an explanatory diagram illustrating an example of short-term data generated from the sales data illustrated in FIG. 3 and the customer data illustrated in FIG. When there is sales that cannot be associated with a customer, the attribute related to the customer may be unknown. In the example shown in FIG. 11, the unknown attribute is represented by “?”. In addition, if there is an attribute that is considered unnecessary for sales, it may be excluded when short-term data is generated by specifying it in advance by the user of the system in order to shorten the calculation time.

  Note that the processing from step S102 to step S110 in FIG. 10 is the same as the processing in the first embodiment, and a description thereof will be omitted.

  Next, the effect of the second embodiment will be described. In the second embodiment, virtual data is generated based on the sales DB 1 and the customer DB 5, and the prediction unit 4 calculates a score of ease of purchase from attributes such as customer characteristics and sales conditions. Therefore, it is possible to analyze and predict which customers are easy to purchase under what sales conditions and when. That is, the demand of individual customers can be predicted.

Embodiment 3. FIG.
The sales number prediction system of the third embodiment is a system having the same effects as those of the first and second embodiments. In particular, since the approximation formula can be simplified, the number of sets of periods used for estimation Suitable for when there is less.

  FIG. 12 is a block diagram showing an example of the third embodiment of the sales number prediction system of the present invention. The sales number prediction system of the third embodiment shown in FIG. 12 includes a sales DB 1, a customer DB 5, a storage unit 7, an analysis unit 8, and a prediction unit 9.

  Sales DB1 and customer DB5 are the same as sales DB1 and customer DB5 in the second embodiment, respectively.

  The storage unit 7 is a virtual DB 71 that stores the attributes and scores of virtual data, a condition DB 22, a classifier DB 23, and an expression for obtaining a threshold value for a sales rate that is a sales ratio of the target product or service in the number of sales. And a threshold DB 72 for storing.

  The analysis unit 8 includes a threshold value smoothing unit 81, a virtual data generation unit 82, and a threshold function estimation unit 83 instead of the time series analysis unit 35 in the analysis unit 6 in the second embodiment.

  The prediction unit 9 includes a virtual short-term data generation unit 91, a score calculation unit 42, a threshold value estimation unit 92, and a sales number prediction unit 44. Hereinafter, only the parts different from the second embodiment will be described.

  The threshold value group smoothing unit 81 smoothes the threshold value estimated by the short-term threshold value group estimating unit 34.

  The virtual data generation unit 82 generates virtual data attributes based on the long-term data generated by the long-term data generation unit 32, and based on the generated virtual data attributes and the classifier stored in the classifier DB 23, Calculate the score of the virtual data. The virtual data is, for example, virtual sales data for calculating a sales rate.

  Based on the score of the virtual data, the threshold function estimating unit 83 estimates an expression for obtaining a threshold Thres [t] at the next time with respect to the sales rate at a certain time, and the estimated formula and the final threshold Thres [ T] is stored in the threshold DB 72.

  The virtual short-term data generation unit 91 generates virtual short-term data attributes based on the virtual data stored in the virtual DB 71.

  The threshold value estimation unit 92 calculates the threshold value at the prediction target time based on the formula for obtaining the threshold value stored in the threshold value DB 72 and the final threshold value Thres [T].

  Next, the operation of the third embodiment will be described with reference to the drawings.

  FIG. 13 is a flowchart showing an example of processing progress of the threshold smoothing unit 81, the virtual data generation unit 82, and the threshold function estimation unit 83 of the analysis unit 8 in the sales number prediction system of the present invention.

  The threshold smoothing unit 81 smoothes Thres [1],... Thres [T] (step S401). Smoothing is performed by taking a moving average or the like. The moving average is to obtain an average of threshold values before and after each threshold value. An expression for obtaining five moving averages for Thres [t] is, for example, as follows.

  (Thres [t−2] + Thres [t−1] + Thres [t] + Thres [t + 1] + Thres [t + 2]) / 5

  The moving average can also be calculated by weighting individual data. A formula for obtaining a 5-day weighted moving average for Thres [t] is, for example, as follows.

  (Thres [t−2] + 2 × Thres [t−1] + 3 × Thres [t] + 2 × Thres [t + 1] + Thres [t + 2]) / (1 + 2 + 3 + 2 + 1)

  Further, outliers of the threshold value may be removed before these processes. Note that the process of step S401 is desirably performed, but is not necessarily required and the process can be omitted.

  The virtual data generation unit 82 generates a virtual data attribute based on the long-term data attribute and stores it in the virtual DB 71 (step S402). Here, the attribute of the virtual data may be the attribute of the long-term data, or an attribute generated by randomly changing some of the attributes of the long-term data may be added.

  Then, the virtual data generation unit 82 calculates the score of the virtual data based on the attribute of the virtual data and the classifier stored in the classifier DB 23 and stores the score in the virtual DB 71 (step S403).

  Next, the virtual data generation unit 82 initializes t = 1 (step S404), and determines the label of the virtual data based on the threshold value Thres [t] (step S405). Then, the virtual data generation unit 82 calculates the sales rate SR [t] (step S406). Here, the sales rate is the number of individual data predicted to be sold with virtual data relative to the number of individual data of virtual data. The virtual data generation unit 82 compares t and T (step S407). If t <T (Yes), t is incremented in step S408, and the operation proceeds to step S405. If t ≧ T (No), the operation moves to step S409.

  FIG. 14 is an explanatory diagram illustrating examples of the score, threshold value, and determination label of virtual data. FIG. 14 shows an example in which the number of sales is only 5 for the sake of explanation. The virtual data generation unit 82 determines that the label is positive (+) if the score of the virtual data is greater than or equal to Thres [t], and the label is negative (−) if the score is less than Thres [t]. To do. When t = 1, the virtual data generation unit 82 calculates the sales rate as SR [1] = 0.2 because the number of individual data with a positive label is 1 and the number of individual data with a negative label is 4. . Similarly, SR [21] = 0.4 and SR [3] = 0.6 are calculated.

  The threshold function estimating unit 83 approximates (SR [t], Thres [t + 1]) (t = 1,..., T−1) from the sales rate SR [t] at a certain time by curve approximation. A threshold function for obtaining a threshold value Thres [t + 1] for determining the presence or absence of sales at the time of is estimated (step S409). The threshold function may be input by the user of the system or may be determined in advance. For example, the threshold function is defined as:

  Thres [t + 1] = a · SR [t] + b

  Here, a and b are parameters to be estimated and can be obtained using a least square method or the like.

  Then, the threshold function estimation unit 83 stores the estimated threshold function and the final threshold Thres [T] in the threshold DB 24 (Step S410).

  FIG. 15 is a flowchart showing an example of processing progress of the prediction unit 9 in the sales number prediction system of the present invention. For example, when the prediction target time TPRED and sales conditions are input via an input device or the like, the following operation is performed. In addition, since there is a track record when TPRED is before T, here, a case where TPRED is after T will be described.

  The virtual short-term data generation unit 91 extracts individual data similar to the input sales conditions from the virtual DB 71, rewrites the sales conditions so as to match the input sales conditions, and generates virtual short-term data ( Step S501).

  The operations in step S202 and step S203 are the same as those in the first and second embodiments.

  The threshold value estimation unit 92 initializes t = T (step S502). Then, the threshold estimation unit 92 obtains the sales rate SR [t] in the virtual data using Thres [t] (step S503). The threshold value estimation unit 92 calculates a threshold value Thres [t + 1] for the sales rate SR [t] based on the formula for obtaining the threshold value stored in the threshold value DB 72 (step S504). Then, t + 1 and TPRED are compared (step S505). If t + 1 <TPRED (Yes), t is incremented (step S506), and the operation is shifted to step S503. If t + 1 ≧ TPRED (No), the operation is moved to step S507.

  In step S507, t + 1 is compared with TPRED, and if t + 1 = TPRED (Yes), the operation moves to step S205, where t + 1! When TPRED (No), Thres [TPRED] is calculated (step S508), and the operation is shifted to step S205. The calculation of Thres [TPRED] is, for example, the average of Thres [t] and Thres [t + 1]. Or you may calculate as follows.

  ((T + 1−TPRED) × Thres [t] + (TPRED−t) × Thres [t + 1])

  Subsequent operations are the same as those in the second embodiment, assuming that the threshold is Thres [TPRED].

  The threshold function estimation unit 83 of the analysis unit 8 calculates the threshold function by curve approximation after calculating the threshold value of the next time in each sales rate. However, the score of the virtual data of the next time in each sales rate is used as it is. The threshold function may be obtained by discriminant analysis. Moreover, although 3rd Embodiment demonstrated the case where there was customer DB5 similarly to 2nd Embodiment, it can also be set as the same structure as 1st Embodiment without customer DB5.

  Also in the third embodiment, the same effect as in the second embodiment can be obtained. In particular, since the threshold value is obtained based on the sales rate related to the threshold value, it can be approximated by a simple expression as compared with the second embodiment in which the threshold value is obtained based on the time, and the number T of sets of periods used for estimation is calculated. The number of sales can be predicted even if there is less.

  In the embodiment described above, a sales number prediction system having characteristic configurations as shown in the following (1) to (5) is shown.

  (1) Based on a sales database (for example, realized by the sales database 1) that stores information indicating one or more sales conditions and past sales of a plurality of products or services, and information stored in the sales database Short-term data that is a set of attributes and labels in a short time series, with the attribute as the sales condition, the label as the first label if one product or service is sold, and the second label if it is not sold Long-term data including all short-term data based on the short-term data group generated by the short-term data group generation unit (for example, realized by the short-term data group generation unit 31) and the short-term data group generation unit. A rule for determining a label based on a long-term data generation unit (for example, realized by the long-term data generation unit 32) and a long-term data attribute; A long-term classifier generating unit (for example, realized by the long-term classifier generating unit 33) that generates a classifier that outputs a score that is a real value indicating the likelihood of the second label or the second label, and each short-term using the classifier A short-term threshold value group estimation unit (for example, a short-term threshold value) that calculates a score of data and calculates a threshold value of a score corresponding to the boundary between the first label and the second label based on the label of each short-term data and the calculated score (Realized by the group estimation unit 34), a time series analysis unit (for example, realized by the time series analysis unit 35) for analyzing a time-series change of the threshold value, sales of long-term data attributes and prediction target time A virtual short-term data generation unit (for example, realized by the virtual short-term data generation unit 41) that generates virtual short-term data including virtual sales conditions as attributes based on the conditions, and a virtual short-term data generation unit Based on the virtual short-term data and the classifier generated by the long-term classifier generator, the score calculator (for example, realized by the score calculator 42) that calculates the score and the result analyzed by the time series analyzer Based on the threshold estimation unit (e.g., realized by the threshold estimation unit 43) that estimates the threshold at the prediction target time, the score calculated by the score calculation unit, and the threshold estimated by the threshold estimation unit A sales number prediction system comprising: a sales number prediction unit (for example, realized by the sales number prediction unit 44) for predicting the sales number at a time.

  (2) A sales database storing information indicating one or more sales conditions and past sales of a plurality of products or services, and an attribute as a sales condition based on the information stored in the sales database, and one label A short-term data group generation unit that generates a set of short-term data that is a set of attributes and labels in a short time series as a first label if a product or service is sold and as a second label if the product or service is not sold ( For example, a short-term data group generation unit 61), a long-term data generation unit that generates long-term data including all short-term data based on a set of short-term data generated by the short-term data group generation unit, Long-term generation of a classifier that determines a label based on an attribute and outputs a score that is a real value indicating the likelihood of a first label or a second label A score generator corresponding to the boundary between the first label and the second label based on the label of each short-term data and the calculated score is calculated using a classifier and a classifier Based on the attributes of the long-term data generated by the short-term threshold group estimation unit and the long-term data generation unit, the virtual data for generating the sales rate including the virtual sales condition as an attribute is generated and the generated virtual The virtual data score is calculated based on the attribute of the data and the classifier generated by the long-term classifier generation unit, and the label of the virtual data is determined based on the calculated score and the threshold value estimated by the short-term threshold group estimation unit. Based on the determined label, the virtual data generation unit (for example, realized by the virtual data generation unit 82) that calculates the sales rate, and the sales rate calculated by the virtual data generation unit and the short-term threshold group estimation unit calculate The virtual sales condition is attributed based on the threshold function estimation unit (e.g., realized by the threshold function estimation unit 83) that estimates the function with the threshold value, the attribute of the virtual data, and the sales condition at the prediction target time. A virtual short-term data generation unit (for example, realized by the virtual short-term data generation unit 91) that generates virtual short-term data for predicting the number of sales included, and virtual short-term data generated by the virtual short-term data generation unit and long-term classification A score calculator that calculates a score based on the classifier generated by the generator generator, and a threshold estimator that estimates a threshold at the prediction target time based on the result of analysis by the time series analyzer (for example, threshold estimation) And a sales number prediction unit that predicts the sales number at the prediction target time based on the score calculated by the score calculation unit and the threshold value estimated by the threshold value estimation unit. Sales number forecasting system characterized by The sales number prediction system thus configured can be approximated by a simple formula, and can predict sales even if the number of sets of periods used for estimation is smaller.

  (3) The short-term data group generation unit, as sales conditions that are attributes, sells products or services, the number of inventory displays, the sales price of other products or services, the number of inventory displays, weather, temperature, temperature changes, days of the week, holidays Sales number forecasting system that generates short-term data including any information indicating.

  (4) A customer database (for example, realized by the customer database 5) that stores information on one or more customers is provided, and the sales database stores information including information that can be associated with customers in part. The short-term data group generation unit is a sales number prediction system that generates short-term data whose attributes are sales conditions and customer information based on information stored in the sales database and the customer database. The sales number prediction system thus configured can analyze the sales conditions and predict the sales number for each customer.

  (5) The customer database includes customer information as attributes, such as customer age, gender, occupation, and most frequently purchased day / time, purchase frequency, and most prominent purchases in various product or service categories. Sales number prediction system for storing customer information including any information indicating day / time, purchase frequency, initial purchase time, and final purchase time.

  The present invention is suitably applied to, for example, an application for predicting the number of retail goods or services sold.

It is a block diagram which shows the minimum structural example of the sales number prediction system by this invention. It is a block diagram which shows the structural example of 1st Embodiment of the sales number prediction system by this invention. It is explanatory drawing which shows the example of the data which sales DB stores. It is explanatory drawing which shows the example of short-term data in the case of estimating the sales number about the goods A. FIG. It is a flowchart which shows the example of the process progress of the analysis part in the sales number prediction system of this invention. It is explanatory drawing which shows the example of the score of the short-term data t. It is a flowchart which shows the example of the process progress of the prediction part in the sales number prediction system of this invention. It is a block diagram which shows the structural example of 2nd Embodiment of the sales number prediction system by this invention. It is explanatory drawing which shows the example of the data which customer DB memorize | stores. It is a flowchart which shows the example of the process progress of the analysis part in the sales number prediction system of this invention. It is explanatory drawing which shows the example of short-term data. It is a block diagram which shows the example of 3rd Embodiment of the sales number prediction system of this invention. It is a flowchart which shows the example of the process progress of the threshold value smoothing part of the analysis part in the sales number prediction system of this invention, a virtual data generation part, and a threshold value function estimation part. It is explanatory drawing which shows the example of the score of virtual data, a threshold value, and a determination label. It is a flowchart which shows the example of the process progress of the prediction part in the sales number prediction system of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Sales database 31 Short-term data group generation part 32 Long-term data generation part 33 Long-term classifier generation part 34 Short-term threshold value group estimation part 35 Time series analysis part 41 Virtual short-term data generation part 42 Score calculation part 43 Threshold estimation part 44 Sales number prediction part

Claims (15)

A sales database storing information indicating one or more sales conditions and past sales of a plurality of goods or services;
Based on the information stored in the sales database, the attribute is the sales condition, the label is the first label if one product or service is sold, and the second label is the second label if it is not sold. A short-term data group generation unit that generates a pair of short-term data that is a set of attributes and labels in
Based on the short-term data set generated by the short-term data group generation unit, a long-term data generation unit that generates long-term data including all short-term data;
A long-term classifier generator that generates a classifier that outputs a score that is a real value indicating the likelihood of the first label or the second label, the rule for determining the label based on the attribute of the long-term data;
A short-term threshold value that calculates a score of each short-term data using the classifier, and calculates a threshold value of a score corresponding to a boundary between the first label and the second label based on the label of each short-term data and the calculated score A group estimator;
A time series analysis unit for analyzing a time series change of the threshold;
A virtual short-term data generation unit that generates virtual short-term data including virtual sales conditions as attributes based on attributes of long-term data and sales conditions at a forecast target time;
A score calculation unit for calculating a score based on the virtual short-term data generated by the virtual short-term data generation unit and the classifier generated by the long-term classifier generation unit;
Based on the analysis result of the time series analysis unit, a threshold estimation unit that estimates a threshold at a prediction target time;
A sales number prediction system comprising: a sales number prediction unit that predicts a sales number at a prediction target time based on the score calculated by the score calculation unit and the threshold value estimated by the threshold value estimation unit. A sales database storing information indicating one or more sales conditions and past sales of a plurality of goods or services;
Based on the information stored in the sales database, the attribute is the sales condition, the label is the first label if one product or service is sold, and the second label is the second label if it is not sold. A short-term data group generation unit that generates a pair of short-term data that is a set of attributes and labels in
Based on the short-term data set generated by the short-term data group generation unit, a long-term data generation unit that generates long-term data including all short-term data;
A long-term classifier generator that generates a classifier that outputs a score that is a real value indicating the likelihood of the first label or the second label, the rule for determining the label based on the attribute of the long-term data;
A short-term threshold value that calculates a score of each short-term data using the classifier, and calculates a threshold value of a score corresponding to a boundary between the first label and the second label based on the label of each short-term data and the calculated score A group estimator;
Based on the long-term data attribute generated by the long-term data generation unit, generates virtual data for calculating a sales rate including a virtual sales condition as an attribute, and generates the generated virtual data attribute and the long-term classifier The virtual data score is calculated based on the classifier generated by the unit, the label of the virtual data is determined based on the calculated score and the threshold value estimated by the short-term threshold value group estimation unit, and based on the determined label A virtual data generation unit for calculating a sales rate;
A threshold function estimation unit that estimates a function of the sales rate calculated by the virtual data generation unit and the threshold value calculated by the short-term threshold group estimation unit;
A virtual short-term data generation unit that generates virtual short-term data for predicting the number of sales including the virtual sales condition as an attribute based on the attribute of the virtual data and the sales condition at the prediction target time;
A score calculation unit for calculating a score based on the virtual short-term data generated by the virtual short-term data generation unit and the classifier generated by the long-term classifier generation unit;
Based on the analysis result of the time series analysis unit, a threshold estimation unit that estimates a threshold at a prediction target time;
A sales number prediction system comprising: a sales number prediction unit that predicts a sales number at a prediction target time based on the score calculated by the score calculation unit and the threshold value estimated by the threshold value estimation unit. The short-term data group generation unit displays the selling price of the product or service, the number of inventory displays, the selling price of other products or services, the number of inventory displays, the weather, temperature, temperature change, day of the week, and holidays. The sales number prediction system according to claim 1 or 2, wherein short-term data including such information is generated. A customer database that stores information for one or more customers;
The sales database stores information including information that can be associated with customers in part,
4. The short-term data group generation unit generates short-term data having attributes as sales conditions and customer information based on information stored in the sales database and the customer database. 5. Sales number forecasting system described in. The customer database includes customer information as attributes, such as customer age, gender, occupation, day / time of purchase most frequently for the entire product, purchase frequency, day of week for most prominent purchase in various product or service categories, The sales number prediction system according to claim 4, storing customer information including any information indicating time, purchase frequency, initial purchase time, and final purchase time. Based on information stored in a sales database that stores information indicating one or more sales conditions and past sales of a plurality of products or services, an attribute is used as the sales conditions and a single product or service is sold. A first label, and if not sold, a second label is generated as a set of short-term data that is a set of attributes and labels in a short time series,
Generate long-term data including all short-term data based on the generated short-term data set,
A classifier that determines a label based on long-term data attributes and outputs a score that is a real value indicating the likelihood of the first label or the second label;
Calculating a score of each short-term data using the classifier, and calculating a threshold of a score corresponding to a boundary between the first label and the second label based on the label of each short-term data and the calculated score;
Analyzing the time series change of the threshold,
Based on the long-term data attributes and the sales conditions at the forecast time, generate virtual short-term data that includes the virtual sales conditions as attributes,
Based on the generated virtual short-term data and the classifier, calculate the score,
Based on the result of analysis, estimate the threshold at the prediction target time,
A sales number prediction method, wherein the number of sales at a prediction target time is predicted based on a calculated score and an estimated threshold value. Based on information stored in a sales database that stores information indicating one or more sales conditions and past sales of a plurality of products or services, an attribute is used as the sales conditions and a single product or service is sold. A first label, and if not sold, a second label is generated as a set of short-term data that is a set of attributes and labels in a short time series,
Generate long-term data including all short-term data based on the generated short-term data set,
A classifier that determines a label based on long-term data attributes and outputs a score that is a real value indicating the likelihood of the first label or the second label;
Calculating a score of each short-term data using the classifier, and calculating a threshold of a score corresponding to a boundary between the first label and the second label based on the label of each short-term data and the calculated score;
Based on the attribute of the generated long-term data, generate virtual data for calculating the sales rate including the virtual sales condition as an attribute, and based on the attribute of the generated virtual data and the generated classifier, Calculate the score, determine the label of the virtual data based on the calculated score and the estimated threshold, calculate the sales rate based on the determined label,
Estimate the function of the calculated sales rate and threshold,
Based on the attributes of the virtual data and the sales conditions at the forecast time, generate virtual short-term data to predict the number of sales that include the virtual sales conditions as attributes,
Based on the generated virtual short-term data and the classifier, calculate the score,
Based on the result of analysis, estimate the threshold at the prediction target time,
A sales number prediction method, wherein the number of sales at a prediction target time is predicted based on a calculated score and an estimated threshold value. Short-term data that includes any information indicating the selling price of the product or service, the number of inventory displays, the selling price of other products or services, the number of inventory displays, weather, temperature, temperature change, day of the week, or holidays as the attribute sales conditions The sales number prediction method according to claim 6 or 7. Based on information stored in a customer database that stores information of one or more customers and a sales database that stores information including information that can be associated with a customer in part, attributes are assigned to sales conditions and customer information. The short-term data used as information is generated. The sales number prediction method according to any one of claims 6 to 8. Customer information includes customer age, gender, profession, most frequently purchased day / time, purchase frequency, most frequently purchased day / time, purchase frequency, initial purchase for various product or service categories The sales number prediction method according to claim 9, wherein short-term data is generated based on information stored in a customer database that stores customer information including any information indicating a time and a final purchase time. On the computer,
Based on information stored in a sales database that stores information indicating one or more sales conditions and past sales of a plurality of products or services, an attribute is used as the sales conditions and a single product or service is sold. A short-term data group generation process for generating a short-term data set that is a set of attributes and labels in a time-series short period as a first label if not sold,
Long-term data generation processing for generating long-term data including all short-term data based on the set of short-term data generated by the short-term data group generation processing;
A rule for determining a label based on an attribute of long-term data, and a long-term classifier generation process for generating a classifier that outputs a score that is a real value indicating the likelihood of the first label or the second label;
A short-term threshold value that calculates a score of each short-term data using the classifier, and calculates a threshold value of a score corresponding to a boundary between the first label and the second label based on the label of each short-term data and the calculated score Group estimation processing;
A time series analysis process for analyzing a time series change of the threshold;
A virtual short-term data generation process for generating virtual short-term data including virtual sales conditions as attributes based on long-term data attributes and sales conditions at the forecast target time;
A score calculation process for calculating a score based on the virtual short-term data generated in the virtual short-term data generation process and the classifier generated in the long-term classifier generation process;
Based on the results analyzed in the time series analysis process, a threshold estimation process for estimating a threshold at the prediction target time;
A sales number prediction program for executing, based on the score calculated in the score calculation process and the threshold value estimated in the threshold value estimation process, a sales number prediction process for predicting the sales number at the prediction target time. On the computer,
Based on information stored in a sales database that stores information indicating one or more sales conditions and past sales of a plurality of products or services, an attribute is used as the sales conditions and a single product or service is sold. A short-term data group generation process for generating a short-term data set that is a set of attributes and labels in a time-series short period as a first label if not sold,
Long-term data generation processing for generating long-term data including all short-term data based on the set of short-term data generated by the short-term data group generation processing;
A rule for determining a label based on an attribute of long-term data, and a long-term classifier generation process for generating a classifier that outputs a score that is a real value indicating the likelihood of the first label or the second label;
A short-term threshold value that calculates a score of each short-term data using the classifier, and calculates a threshold value of a score corresponding to a boundary between the first label and the second label based on the label of each short-term data and the calculated score Group estimation processing;
Based on the long-term data attribute generated in the long-term data generation process, generates virtual data for calculating a sales rate including a virtual sales condition as an attribute, and generates the generated virtual data attribute and the long-term classifier The virtual data score is calculated based on the classifier generated by the process, the label of the virtual data is determined based on the calculated score and the threshold value estimated in the short-term threshold value group estimation process, and based on the determined label Virtual data generation processing to calculate the sales rate,
A threshold function estimation process for estimating a function of the sales rate calculated in the virtual data generation process and the threshold value calculated in the short-term threshold group estimation process;
Virtual short-term data generation processing for generating virtual short-term data for predicting the number of sales including the virtual sales conditions as attributes based on the attributes of the virtual data and the sales conditions at the prediction target time;
A score calculation process for calculating a score based on the virtual short-term data generated in the virtual short-term data generation process and the classifier generated in the long-term classifier generation process;
Based on the results analyzed in the time series analysis process, a threshold estimation process for estimating a threshold at the prediction target time;
A sales number prediction program for executing, based on the score calculated in the score calculation process and the threshold value estimated in the threshold value estimation process, a sales number prediction process for predicting the sales number at the prediction target time. In the short-term data group generation process, the sales conditions that are the attributes include the selling price of goods or services, the number of inventory displays, the selling price of other products or services, the number of inventory displays, weather, temperature, temperature change, day of the week, and holidays The sales number prediction program according to claim 11 or 12, wherein short-term data including such information is generated. Based on information stored in a customer database that stores information of one or more customers and a sales database that stores information including information that can be associated with a customer in part, attributes are assigned to sales conditions and customer information. The short-term data as information is generated. The sales number prediction program according to any one of claims 11 to 13. Customer information includes customer age, gender, profession, most frequently purchased day / time, purchase frequency, most frequently purchased day / time, purchase frequency, initial purchase for various product or service categories The sales number prediction program according to claim 14, wherein short-term data is generated based on information stored in a customer database that stores customer information including any information indicating a time and a final purchase time.

Images